diff --git a/lib/hwasan/hwasan_tag_mismatch_aarch64.S b/lib/hwasan/hwasan_tag_mismatch_aarch64.S
index 92f6274..4c060a6 100644
--- a/lib/hwasan/hwasan_tag_mismatch_aarch64.S
+++ b/lib/hwasan/hwasan_tag_mismatch_aarch64.S
@@ -51,14 +51,60 @@
 // +---------------------------------+ <-- [x30 / SP]
 
 // This function takes two arguments:
-//   * x0: The address of read/write instruction that caused HWASan check fail.
-//   * x1: The tag size.
+//   * x0: The data address.
+//   * x1: The encoded access info for the failing access.
 
+// This function has two entry points. The first, __hwasan_tag_mismatch, is used
+// by clients that were compiled without short tag checks (i.e. binaries built
+// by older compilers and binaries targeting older runtimes). In this case the
+// outlined tag check will be missing the code handling short tags (which won't
+// be used in the binary's own stack variables but may be used on the heap
+// or stack variables in other binaries), so the check needs to be done here.
+//
+// The second, __hwasan_tag_mismatch_v2, is used by binaries targeting newer
+// runtimes. This entry point bypasses the short tag check since it will have
+// already been done as part of the outlined tag check. Since tag mismatches are
+// uncommon, there isn't a significant performance benefit to being able to
+// bypass the check; the main benefits are that we can sometimes avoid
+// clobbering the x17 register in error reports, and that the program will have
+// a runtime dependency on the __hwasan_tag_mismatch_v2 symbol therefore it will
+// fail to start up given an older (i.e. incompatible) runtime.
 .section .text
 .file "hwasan_tag_mismatch_aarch64.S"
 .global __hwasan_tag_mismatch
 .type __hwasan_tag_mismatch, %function
 __hwasan_tag_mismatch:
+  // Compute the granule position one past the end of the access.
+  mov x16, #1
+  and x17, x1, #0xf
+  lsl x16, x16, x17
+  and x17, x0, #0xf
+  add x17, x16, x17
+
+  // Load the shadow byte again and check whether it is a short tag within the
+  // range of the granule position computed above.
+  ubfx x16, x0, #4, #52
+  ldrb w16, [x9, x16]
+  cmp w16, #0xf
+  b.hi __hwasan_tag_mismatch_v2
+  cmp w16, w17
+  b.lo __hwasan_tag_mismatch_v2
+
+  // Load the real tag from the last byte of the granule and compare against
+  // the pointer tag.
+  orr x16, x0, #0xf
+  ldrb w16, [x16]
+  cmp x16, x0, lsr #56
+  b.ne __hwasan_tag_mismatch_v2
+
+  // Restore x0, x1 and sp to their values from before the __hwasan_tag_mismatch
+  // call and resume execution.
+  ldp x0, x1, [sp], #256
+  ret
+
+.global __hwasan_tag_mismatch_v2
+.type __hwasan_tag_mismatch_v2, %function
+__hwasan_tag_mismatch_v2:
   CFI_STARTPROC
 
   // Set the CFA to be the return address for caller of __hwasan_check_*. Note
diff --git a/test/hwasan/TestCases/stack-oob.c b/test/hwasan/TestCases/stack-oob.c
index ba74930..8c8c110 100644
--- a/test/hwasan/TestCases/stack-oob.c
+++ b/test/hwasan/TestCases/stack-oob.c
@@ -1,4 +1,8 @@
+// RUN: %clang_hwasan_oldrt -DSIZE=2 -O0 %s -o %t && %run %t
+// RUN: %clang_hwasan -DSIZE=2 -O0 %s -o %t && not %run %t 2>&1 | FileCheck %s
+// RUN: %clang_hwasan_oldrt -DSIZE=15 -O0 %s -o %t && %run %t
 // RUN: %clang_hwasan -DSIZE=15 -O0 %s -o %t && not %run %t 2>&1 | FileCheck %s
+// RUN: %clang_hwasan_oldrt -DSIZE=16 -O0 %s -o %t && not %run %t 2>&1 | FileCheck %s
 // RUN: %clang_hwasan -DSIZE=16 -O0 %s -o %t && not %run %t 2>&1 | FileCheck %s
 // RUN: %clang_hwasan -DSIZE=64 -O0 %s -o %t && not %run %t 2>&1 | FileCheck %s
 // RUN: %clang_hwasan -DSIZE=0x1000 -O0 %s -o %t && not %run %t 2>&1 | FileCheck %s
@@ -16,9 +20,9 @@
 }
 
 int main() {
-  return f();
+  f();
   // CHECK: READ of size 1 at
-  // CHECK: #0 {{.*}} in f{{.*}}stack-oob.c:15
+  // CHECK: #0 {{.*}} in f{{.*}}stack-oob.c:19
 
   // CHECK: is located in stack of threa
 
diff --git a/test/hwasan/lit.cfg.py b/test/hwasan/lit.cfg.py
index eead832..ee67a26 100644
--- a/test/hwasan/lit.cfg.py
+++ b/test/hwasan/lit.cfg.py
@@ -18,9 +18,11 @@
   # equivalent target feature implemented on x86_64.
   clang_hwasan_common_cflags += ["-mcmodel=large"]
 clang_hwasan_cflags = clang_hwasan_common_cflags + ["-mllvm", "-hwasan-globals",
+                                                   "-mllvm", "-hwasan-use-short-granules",
                                                    "-mllvm", "-hwasan-instrument-landing-pads=0",
                                                    "-mllvm", "-hwasan-instrument-personality-functions"]
-clang_hwasan_oldrt_cflags = clang_hwasan_common_cflags + ["-mllvm", "-hwasan-instrument-landing-pads=1",
+clang_hwasan_oldrt_cflags = clang_hwasan_common_cflags + ["-mllvm", "-hwasan-use-short-granules=0",
+                                                          "-mllvm", "-hwasan-instrument-landing-pads=1",
                                                           "-mllvm", "-hwasan-instrument-personality-functions=0"]
 
 clang_hwasan_cxxflags = config.cxx_mode_flags + clang_hwasan_cflags
@@ -31,6 +33,7 @@
 
 config.substitutions.append( ("%clangxx ", build_invocation(clang_cxxflags)) )
 config.substitutions.append( ("%clang_hwasan ", build_invocation(clang_hwasan_cflags)) )
+config.substitutions.append( ("%clang_hwasan_oldrt ", build_invocation(clang_hwasan_oldrt_cflags)) )
 config.substitutions.append( ("%clangxx_hwasan ", build_invocation(clang_hwasan_cxxflags)) )
 config.substitutions.append( ("%clangxx_hwasan_oldrt ", build_invocation(clang_hwasan_oldrt_cxxflags)) )
 config.substitutions.append( ("%compiler_rt_libdir", config.compiler_rt_libdir) )